rajneeshsoftware developer · coding since 2014
i'm rajneesh. i build software.the work lives inside.
$ whoami
software developer · coding since 2014· 100+ projects· 7 ai systems· 40+ clients
$ ls ~/work
health-intelligence/herzify-model/friend-bot/shieldy/skillgen/shorts-studio/takeproof/vle-platform/choreless/…and 80 more
$ ./enter
↑ that's a lens, not a camera. nothing is recorded.
- raj, the one in white. rajneesh. coding since 2014: frontends, backends, automation and ai. bca, mca. leads projects at grapdevs.com, makes music when the code is quiet.
- health intelligence, 2026 · product architect. medical reports sit in pdfs, prescriptions get ignored, patterns go unnoticed. a system that reads the reports, learns long-term patterns, scores risk early and explains it in plain language. Inside it: ocr + nlp extraction, medical term recognition, structured json records, feature engineering, pattern learning model, risk scoring engine, drug interaction checks, adaptive reminders, llm health assistant, doctor-ready summaries, aes-256 + tls 1.3, pii anonymization, audit logging. Built with: next.js, react, tailwind, framer motion, python, llms · gpt · ollama, nlp, ocr.
- herzify model, 2026 · ai model. the ai behind herzify: it learns how you feel and how you listen, then picks what you should hear next. mood in, playlist out, running on gpt or a local ollama model. Inside it: mood → playlist engine, listening-pattern learning, mood-based recommendation, spotify integration, gpt + ollama, midi playlists (next). Built with: next.js, llms · gpt · ollama, spotify api, firebase.
- friend bot, 2026 · conversational ai. most ai replies and forgets. this one is being built to remember: continuity, emotional tone, a personality you choose. not a chatbot. a friend that grows with you. Inside it: long-term memory, emotional tone tracking, personality you choose, learns your writing style, private reflection space, training & research. Built with: llms · gpt · ollama, nlp, python.
- shieldy, 2026 · ai memory for client work. a client-centric ai memory for company communications. it connects gmail, whatsapp, files and figma, matches every message to the right client, turns it into searchable memory and answers anything about an account. Inside it: gmail · whatsapp · files · figma, matches messages to clients, vector memory, ask anything about a client, brief replies, notion tasks, composio integrations, agent evals · head to head, self-healing jobs, usage metering, super-admin console. Built with: laravel, php / codeigniter, react, tailwind, express, llms · gpt · ollama.
- skillgen, 2026 · open source · pypi. a compiler that reads an app's code, route by route, and writes one agent skill per page, so claude or any agent can fill that page's forms in one browser call instead of clicking field by field. Inside it: finds every route, reads the code behind each form, one SKILL.md per page, scripts/fill.js, framework-agnostic, browser probe check, few-token form filling, pypi: skillgendev. Built with: python, claude · mcp · agent skills, llms · gpt · ollama.
- shorts studio, 2026 · local ai pipeline. a local, low-cost youtube shorts factory. paste a link: it downloads, transcribes, finds the strongest moments, cuts them to 9:16 with face tracking, burns in captions, writes titles and tags and suggests when to post. Inside it: download any video (yt-dlp), faster-whisper transcription, picks the strongest moments, 9:16 face-tracked cuts, burned-in captions, titles & tags by a model, best time to post, tracks every clip to live. Built with: python, fastapi / flask, llms · gpt · ollama, opencv, ffmpeg, docker.
- takeproof, 2026 · local-first windows app. for music producers: it records how a track is made in fl studio and turns that record into an honest, per-element ai disclosure and a proof pack anyone can verify with a free tool. Inside it: fl studio session capture, per-element ai disclosure, verifiable proof pack, free verifier, local-first, private. Built with: python, docker.
- llm reasoning & prompting. reasoning over structured context, not guessing: plain-language explanations and doctor-ready summaries.
- local models · ollama. running models on your own machine as well as through apis.
- ocr pipelines. turning scanned reports, pdfs and lab layouts into text a system can use.
- medical entity recognition. finding drugs, values and conditions inside messy medical text, then normalising them to json.
- risk scoring & anomaly detection. longitudinal tracking that flags sudden changes and rising risk early.
- learning from behaviour. adherence patterns, listening patterns: systems that adapt to what people actually do.
- mood-based recommendation. matching music to how someone feels and how they listen.
- conversational memory. continuity across conversations: what was shared does not disappear.
- tone & personality control. calm, direct, motivating or just listening: behaviour you choose, that evolves.
- privacy before inference. pii stripped before anything reaches a model; encrypted at rest and in transit; every access logged.
- ai in operations. marketing and ai tools inside a real business, for the people doing the work.
- statistical ranking. percentiles, normalisation and tie-breaks that stay fair at exam scale.
- agent skills & mcp. teaching agents an app once: skills that fill a page in one call, mcp servers that plug claude into real tools.
- rag & vector memory. matching messages to the right client, embedding them, and answering from that memory.
- agent evaluation. head-to-head and multi-turn evals, reliability runs and scorecards before an agent ships.
- provider-agnostic ai. one switch between model vendors (nvidia nim, openrouter, any openai-compatible api), so a product never depends on one.
- ai seo & content. keywords, titles, meta descriptions, tags and review summaries written by models, inside real products.
- speech & video ai. whisper transcription, picking the strongest moments, face-tracked 9:16 cuts and burned-in captions.
- ai provenance & disclosure. recording how a track was made, so its use of ai can be disclosed honestly and verified by anyone.
- training models · pytorch. scraping a corpus, training a model and serving it behind a web interface.
- vle platform, full stack architect. syllabus tracking, online exams, ranking and analytics in one place for coaching centres. camera and audio checks keep exams honest; python does the ranking. Inside it: role-based auth, syllabus hierarchy & coverage, timed sectional exams, negative marking, session lock + auto submit, camera + audio proctoring, python ranking engine, percentiles & tie-breaks, shared notes, dashboards & reports. Built with: next.js, php / codeigniter, python, docker, jenkins, sql.
- choreless, 2020–23 · full stack. a laundry business, end to end: washers, orders, quality, admin panels, web apps, an order-management app, and the marketing and ai tools that tie it together. Inside it: washer management, order & processing layout, quality checks, admin panels, order-management app, websites & web apps, marketing + ai tools. Built with: vue / nuxt, tailwind, next.js, react, laravel, php / codeigniter, node.js, sql, firebase, aws, cloudflare, wordpress.
- water harvesting, 2021–22 · front-end & api. an iot dashboard for urban water harvesting: sensors on a raspberry pi, savings per zone, and a ranking of where the water can be reused. Inside it: raspberry pi sensors, zones: bath, garden, pool, water-savings calculator, reuse suitability ranking, live dashboard. Built with: raspberry pi, vue / nuxt, react, python, mongodb, firebase.
- tv app, 2023 · ui/ux & front-end. a film and tv streaming app: aws transcoding, channels, movies and shows over a cdn, subtitles during playback, and subscriptions. Inside it: aws transcoder pipeline, channels, movies & shows over cdn, live subtitles, subscriptions, tv-first interface. Built with: aws, react, react native, stripe, adobe xd.
- author dashboard, 2022 · rosewood publications. authors publish books, track earnings from amazon and flipkart, and manage their copies in one place. Inside it: book publishing, publication management, amazon + flipkart earnings, author copies, ai sales analysis. Built with: firebase, bootstrap, adobe xd.
- restaurant pos, 2026 · full stack. a real-time point of sale for restaurants: a next.js front, an express api, and every screen kept live over websockets. Inside it: live updates over socket.io, next.js admin, express api, sql database. Built with: next.js, react, typescript, tailwind, express, node.js, socket.io, sql.
- bugspoon, 2025 · chrome extension & devtools. a devtools panel that records console logs and network traces, so every bug report arrives with its own evidence. built on manifest v3, fast at 1000+ requests a minute. Inside it: network+ panel, console log capture, zero-latency filtering, service-worker isolation, instant replay, long-lived port relay, relay server (laravel), node cli, shareable bug reports, alerts. Built with: javascript, chrome extension apis, react, tailwind, framer motion, laravel, php / codeigniter, vue / nuxt, express, node.js.
- oggo, 2026 · self-hosted ops. a self-hosted cron job and server-management platform with a built-in ssh terminal. it started as my own control centre and is open source now. Inside it: cron jobs across servers, built-in ssh terminal, aws: ec2, lambda, rds, s3, ses, job monitoring, self-hosted. Built with: node.js, express, javascript, sql, aws.
- gitsmith, npm · ai commits cli. ai-assisted conventional commits for teams: a cli that keeps commit messages consistent, suggests them with a model, and runs safety checks before it commits. Inside it: .commitconfig.json formats, ai suggestions (nvidia), safety checks, manual fallback. Built with: node.js, javascript, nvidia nim.
- life os, claude ↔ notion. an mcp server that connects claude to your life systems, starting with notion: tasks, notes and plans, managed by talking. Inside it: mcp server, notion tasks & notes, claude tools. Built with: typescript, node.js, claude · mcp · agent skills, notion api.
- ai lyrics generator, pytorch · end to end. scrapes a lyrics corpus, trains a pytorch model, and serves a web interface that writes new lyrics from a prompt. Inside it: corpus scraper, pytorch training, prompt → lyrics, web interface. Built with: python, pytorch, nlp, fastapi / flask.
- youtube → daw cues, for producers. capture cue points from any youtube video and export them as a midi file: every cue becomes a note in your daw. Inside it: cue capture on youtube, midi export, drops into any daw. Built with: chrome extension apis, javascript, midi.
- cachemate, react · next.js. offline-first axios caching for react and next.js: indexeddb with a localstorage fallback, and a queue that replays writes when you are back online. Inside it: indexeddb cache, localstorage fallback, offline mutation queue. Built with: react, next.js, javascript.
- ghostlink, local reverse proxy. a windows-friendly local reverse proxy: map the apps on your machine to friendly .local domains across your lan. Inside it: .local domains, reverse proxy, lan routing. Built with: node.js, javascript.
- flowterminal, electron · react. a modern, visual terminal application for the desktop. Inside it: visual terminal, desktop app. Built with: electron, react, typescript, tailwind.
- react-nextjs-github-gist, npm · 900+ downloads a week. embed github gists in react and next.js, with the author's name and avatar. Inside it: gist embeds, author name + avatar, client-side rendering. Built with: react, next.js, typescript.
- breww scroll-spyer, 2023 · npm. a scrollspy library with animations, written in my free time. Inside it: scrollspy, section animations, npm i @breww.io/scroll-spyer. Built with: javascript, typescript, react.
- is-business-mail, npm. tells a work address from a gmail one. small, boring, useful. Built with: typescript, javascript.
- code dictionary, snippets · ctrl+c, ctrl+v. code samples worth keeping, like protecting routes with next 14 middleware. Built with: next.js, react, typescript, tailwind, sanity.
- pentium 4, 2008. fourth grade. the first pc: windows xp, green wallpaper, that startup sound. the whole neighbourhood came to watch it boot.
- command prompt, 5th grade. my father, an army man, brought home a book about it. a black screen, one white cursor. a secret language.
- first webpage, 6th grade. a senior handed me his html notes. i typed a few lines. it worked. i was gone.
- first real projects, 2014. php, html and css, built for real while still studying (bca, then mca). full stack, the hard way.
- building for clients, 2020. built and sold platforms to clients. learned how to ship under pressure.
- iot & scale, 2021–22. water-harvesting dashboards on raspberry pis, and big platforms like vle.
- apps & cloud, 2023. tv streaming apps on aws transcoder stacks.
- tooling, 2024–25. bugspoon, and open-source libraries other developers install every week.
- ai, music & vision, 2026. health intelligence, shieldy, skillgen, shorts studio, takeproof, herzify model, friend bot: backend, frontend, ai and music in one era.
- grapdevs, now. where i lead projects today.
- jeff.co. one of the teams i've built with.
- karat.io. one of the teams i've built with.
- mithi. one of the teams i've built with.
- petofy. one of the teams i've built with.
- cynotech. one of the teams i've built with.
- fusioni tech. one of the teams i've built with.
- rosewood. one of the teams i've built with.
- choreless. one of the teams i've built with.
- marketing crawlers. one of the teams i've built with.
- prm digital. one of the teams i've built with.
- truedreamster. one of the teams i've built with.
- legacyflow. one of the teams i've built with.
- wibyng. one of the teams i've built with.
- kapiva. one of the teams i've built with.
- fm1. one of the teams i've built with.
- dreamershelf. one of the teams i've built with.
- the fold. one of the teams i've built with.
- wapka. one of the teams i've built with.
- educrafter. one of the teams i've built with.
- rayforce greentech. one of the teams i've built with.
- haitinflix. one of the teams i've built with.
- motosports. one of the teams i've built with.
- jflex. one of the teams i've built with.
- next.js
- react
- typescript
- javascript
- node.js
- express
- python
- fastapi / flask
- php / codeigniter
- laravel
- vue / nuxt
- tailwind
- bootstrap
- framer motion
- llms · gpt · ollama
- claude · mcp · agent skills
- nvidia nim
- pytorch
- nlp
- ocr
- opencv
- ffmpeg
- chrome extension apis
- electron
- socket.io
- raspberry pi
- react native
- firebase
- mongodb
- sql
- sanity
- notion api
- aws
- cloudflare
- docker
- jenkins
- stripe
- spotify api
- midi
- wordpress
- adobe xd
- writing
- The six layers I keep building into every AI system
- From a scanned PDF to clean JSON an AI can trust
- Giving an AI long-term memory with vectors and RAG
- Privacy before inference: strip personal data before the AI sees it
- Never depend on one AI provider: a small layer that switches models
- Teaching AI agents to fill a whole form in one call
- Testing AI agents before you ship them
- Turning a long video into shorts with a local AI pipeline
- Spotting risk over time: let code do the maths and AI do the explaining
- Making your website readable by AI agents with llms.txt
- "I'll help you out" vs "I'll clarify this for you." An engine that knows the difference.
- Scaled Server for High Traffic App around 25k hits daily
- All about RDBMS (SQL) | Benefits and limitations of SQL & NoSQL
- Share Anything with native share options in Html & JavaScript
- Leveraging Jenkins with Private GitHub and EC2 for Seamless Development Workflow